Repricing rules that cannot quietly break your margin

A repricing rule has five parts: the products it covers, the competitor listings it is allowed to read, the target it aims for, the floors it can never cross, and where the resulting price is sent. The floors are the part that matters. A rule without a margin floor, a cost floor and a MAP floor is a way to lose money quickly and consistently, and every rule here carries all three.

What repricing rules are made of

Rules are built from named parts rather than a formula box, so anyone on the team can read a rule six months later and know what it does.

Guardrails: the floors a rule can never cross

Guardrails are evaluated after the target, and the most restrictive one wins. That is the entire safety model, and it is worth walking through with numbers because it is where most repricing accidents happen.

Take a product currently listed at $89.99 with a landed cost of $62.00. The lowest in-stock matched competitor is $84.99, and the rule says to beat the lowest by a cent, so the target price is $84.98. Four constraints now apply.

That last detail matters more than it sounds. A rule that silently fails to reach its target looks identical to a rule that is working, until margin erodes somewhere else. Every proposal carries the constraint that bound it. If a whole category is repeatedly held by its margin floor, the answer is usually that you are not competitive on cost in that category, and no repricing rule fixes that.

How repricing consumes matched competitor data

A repricing rule is only as good as the observations feeding it, so the rule engine is deliberately picky about what it will read. Data has to clear five filters before it can move a price.

1. 1Match confidence. Only matches above the confidence threshold you set are eligible. Low-confidence candidates never move a price, they go to review. The waterfall behind those scores is described in the product matching guide.
2. 2Stock status. Out-of-stock listings are excluded by default. Chasing a price nobody can buy is the most common way an automated rule gives away margin for nothing.
3. 3Freshness. An observation older than the window you set is ignored. If a source has not returned a good price for two refresh cycles, its last known price stops counting rather than quietly anchoring the rule.
4. 4Validation. Prices quarantined by plausibility checks never reach a rule. A $0.00 reading or a shipping fee misread as a price is held back at collection, which is covered in price monitoring.
5. 5Seller filters. You can restrict a rule to authorized retailers, exclude specific marketplace sellers, or exclude anyone currently in violation, so a rule does not chase a price that MAP protection is actively enforcing against.

Variants are priced individually. A rule evaluates the 5Ah battery against 5Ah listings, not against the 2Ah on the same competitor page, and pack sizes are compared on unit price where they differ.

Review queue, CSV export, or push to your store

Every rule sends its output to exactly one destination, chosen per rule. Auto-push is off by default on every account and every rule. Nothing is written to a connected store until you turn that on for a specific rule, and turning it off again takes one click.

Push results are recorded, not assumed. If a store rejects a write, the change is marked failed with the error returned, and it is not counted as applied. A rule cannot report success for a price that never landed.

Change history and rollback

Every change is logged, whether it was proposed, approved, rejected, pushed, failed or rolled back. History is append-only, so nothing in it can be edited after the fact.

• Timestamp, and whether the actor was a rule or a named user.
• The rule that produced it, its target and its guardrail settings at that moment, so a later edit to the rule does not rewrite the reason for an old change.
• Product and variant, old price, new price, and the change in dollars and percent.
• The competitor observations that triggered it, each with its source URL and capture time.
• The constraint that bound the price, where the rule did not reach its target.
• The destination and the result, including any error a store returned.

Rollback restores the previous price for one change or a whole batch, and it is written as a new entry rather than erasing the original. That matters when someone asks in October why a price moved in July: the record shows both the move and the reversal, with the reasoning attached to each. Rules can also be paused, which stops new proposals without discarding the queue.

A rollout that does not frighten anyone

1. 1

   Week one, watch only

   Point a rule at one category with the review queue as its destination, and approve nothing. Read the proposals daily. You are testing whether the rule reaches sensible numbers, not repricing yet.

2. 2

   Week two, approve in batches

   Approve the obvious ones, reject the rest, and note why. Rejection reasons almost always point at a filter that needs tightening rather than a target that needs changing.

3. 3

   Week three, narrow the guardrails

   Set the maximum daily change to something conservative, 3 to 5 percent is a common starting point, and confirm the binding constraint on rejected proposals is one you agree with.

4. 4

   Week four, enable push on a subset

   Turn on store push for a defined group of products where the rule has been right consistently. Keep the review queue for everything else. Expand only when the change history gives you a reason to.

Strategy choices, such as when to match, when to hold and when to ignore a competitor entirely, are covered in repricing strategies. This page is about the machinery.